One of the primary public wellness concerns is Chronic Kidney Disease (CKD) that needs very early detection, administration for efficient get rid of. This paper aims to analyze the performance of K-Nearest Neighbours (KNN) algorithm for diagnosing Chronic Kidney Disease (CKD) and investigates a series of preprocessing methods that may have an impact on model prediction scores. The dataset consists of 400 samples with 25 features in CKD dataset which is a compilation of clinical and laboratory data. The use of three preprocessing techniques: Min-Max scaling, boxplot analysis to detect and remove outliers using Z-score normalization. The performance or the outcome of model is done using three measures i.e., Log Loss (LL), Balanced Accuracy (BA) and Matthews Correlation Coefficient (MCC). The results also indicate that KNN with the boxplot preprocessing provide significant better performance compared to other alternative preprocessors. Smoothing over the outliers and scaling properly results in improved MCC, Log Loss reduction, and balanced accuracy. As well as showcasing KNN accompanied by a boxplot analysis to be an effective tool in accurate diagnosis (and perhaps for aiding medical professionals with early recognition and intervention), this study highlights the importance of preprocessing tailored towards CKD risk prediction.